1 #include <linux/workqueue.h>
2 #include <linux/rtnetlink.h>
3 #include <linux/cache.h>
4 #include <linux/slab.h>
5 #include <linux/list.h>
6 #include <linux/delay.h>
7 #include <linux/sched.h>
9 #include <linux/rculist.h>
10 #include <linux/nsproxy.h>
11 #include <linux/proc_fs.h>
12 #include <linux/file.h>
13 #include <net/net_namespace.h>
14 #include <net/netns/generic.h>
17 * Our network namespace constructor/destructor lists
20 static LIST_HEAD(pernet_list);
21 static struct list_head *first_device = &pernet_list;
22 static DEFINE_MUTEX(net_mutex);
24 LIST_HEAD(net_namespace_list);
25 EXPORT_SYMBOL_GPL(net_namespace_list);
28 EXPORT_SYMBOL(init_net);
30 #define INITIAL_NET_GEN_PTRS 13 /* +1 for len +2 for rcu_head */
32 static int net_assign_generic(struct net *net, int id, void *data)
34 struct net_generic *ng, *old_ng;
36 BUG_ON(!mutex_is_locked(&net_mutex));
39 old_ng = rcu_dereference_protected(net->gen,
40 lockdep_is_held(&net_mutex));
42 if (old_ng->len >= id)
45 ng = kzalloc(sizeof(struct net_generic) +
46 id * sizeof(void *), GFP_KERNEL);
51 * Some synchronisation notes:
53 * The net_generic explores the net->gen array inside rcu
54 * read section. Besides once set the net->gen->ptr[x]
55 * pointer never changes (see rules in netns/generic.h).
57 * That said, we simply duplicate this array and schedule
58 * the old copy for kfree after a grace period.
62 memcpy(&ng->ptr, &old_ng->ptr, old_ng->len * sizeof(void*));
64 rcu_assign_pointer(net->gen, ng);
65 kfree_rcu(old_ng, rcu);
67 ng->ptr[id - 1] = data;
71 static int ops_init(const struct pernet_operations *ops, struct net *net)
74 if (ops->id && ops->size) {
75 void *data = kzalloc(ops->size, GFP_KERNEL);
79 err = net_assign_generic(net, *ops->id, data);
86 return ops->init(net);
90 static void ops_free(const struct pernet_operations *ops, struct net *net)
92 if (ops->id && ops->size) {
94 kfree(net_generic(net, id));
98 static void ops_exit_list(const struct pernet_operations *ops,
99 struct list_head *net_exit_list)
103 list_for_each_entry(net, net_exit_list, exit_list)
107 ops->exit_batch(net_exit_list);
110 static void ops_free_list(const struct pernet_operations *ops,
111 struct list_head *net_exit_list)
114 if (ops->size && ops->id) {
115 list_for_each_entry(net, net_exit_list, exit_list)
121 * setup_net runs the initializers for the network namespace object.
123 static __net_init int setup_net(struct net *net)
125 /* Must be called with net_mutex held */
126 const struct pernet_operations *ops, *saved_ops;
128 LIST_HEAD(net_exit_list);
130 atomic_set(&net->count, 1);
131 atomic_set(&net->passive, 1);
133 #ifdef NETNS_REFCNT_DEBUG
134 atomic_set(&net->use_count, 0);
137 list_for_each_entry(ops, &pernet_list, list) {
138 error = ops_init(ops, net);
146 /* Walk through the list backwards calling the exit functions
147 * for the pernet modules whose init functions did not fail.
149 list_add(&net->exit_list, &net_exit_list);
151 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
152 ops_exit_list(ops, &net_exit_list);
155 list_for_each_entry_continue_reverse(ops, &pernet_list, list)
156 ops_free_list(ops, &net_exit_list);
162 static struct net_generic *net_alloc_generic(void)
164 struct net_generic *ng;
165 size_t generic_size = sizeof(struct net_generic) +
166 INITIAL_NET_GEN_PTRS * sizeof(void *);
168 ng = kzalloc(generic_size, GFP_KERNEL);
170 ng->len = INITIAL_NET_GEN_PTRS;
176 static struct kmem_cache *net_cachep;
177 static struct workqueue_struct *netns_wq;
179 static struct net *net_alloc(void)
181 struct net *net = NULL;
182 struct net_generic *ng;
184 ng = net_alloc_generic();
188 net = kmem_cache_zalloc(net_cachep, GFP_KERNEL);
192 rcu_assign_pointer(net->gen, ng);
201 static void net_free(struct net *net)
203 #ifdef NETNS_REFCNT_DEBUG
204 if (unlikely(atomic_read(&net->use_count) != 0)) {
205 printk(KERN_EMERG "network namespace not free! Usage: %d\n",
206 atomic_read(&net->use_count));
211 kmem_cache_free(net_cachep, net);
214 void net_drop_ns(void *p)
217 if (ns && atomic_dec_and_test(&ns->passive))
221 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
226 if (!(flags & CLONE_NEWNET))
227 return get_net(old_net);
231 return ERR_PTR(-ENOMEM);
232 mutex_lock(&net_mutex);
236 list_add_tail_rcu(&net->list, &net_namespace_list);
239 mutex_unlock(&net_mutex);
247 static DEFINE_SPINLOCK(cleanup_list_lock);
248 static LIST_HEAD(cleanup_list); /* Must hold cleanup_list_lock to touch */
250 static void cleanup_net(struct work_struct *work)
252 const struct pernet_operations *ops;
253 struct net *net, *tmp;
254 LIST_HEAD(net_kill_list);
255 LIST_HEAD(net_exit_list);
257 /* Atomically snapshot the list of namespaces to cleanup */
258 spin_lock_irq(&cleanup_list_lock);
259 list_replace_init(&cleanup_list, &net_kill_list);
260 spin_unlock_irq(&cleanup_list_lock);
262 mutex_lock(&net_mutex);
264 /* Don't let anyone else find us. */
266 list_for_each_entry(net, &net_kill_list, cleanup_list) {
267 list_del_rcu(&net->list);
268 list_add_tail(&net->exit_list, &net_exit_list);
273 * Another CPU might be rcu-iterating the list, wait for it.
274 * This needs to be before calling the exit() notifiers, so
275 * the rcu_barrier() below isn't sufficient alone.
279 /* Run all of the network namespace exit methods */
280 list_for_each_entry_reverse(ops, &pernet_list, list)
281 ops_exit_list(ops, &net_exit_list);
283 /* Free the net generic variables */
284 list_for_each_entry_reverse(ops, &pernet_list, list)
285 ops_free_list(ops, &net_exit_list);
287 mutex_unlock(&net_mutex);
289 /* Ensure there are no outstanding rcu callbacks using this
294 /* Finally it is safe to free my network namespace structure */
295 list_for_each_entry_safe(net, tmp, &net_exit_list, exit_list) {
296 list_del_init(&net->exit_list);
300 static DECLARE_WORK(net_cleanup_work, cleanup_net);
302 void __put_net(struct net *net)
304 /* Cleanup the network namespace in process context */
307 spin_lock_irqsave(&cleanup_list_lock, flags);
308 list_add(&net->cleanup_list, &cleanup_list);
309 spin_unlock_irqrestore(&cleanup_list_lock, flags);
311 queue_work(netns_wq, &net_cleanup_work);
313 EXPORT_SYMBOL_GPL(__put_net);
315 struct net *get_net_ns_by_fd(int fd)
317 struct proc_inode *ei;
321 file = proc_ns_fget(fd);
323 return ERR_CAST(file);
325 ei = PROC_I(file->f_dentry->d_inode);
326 if (ei->ns_ops == &netns_operations)
327 net = get_net(ei->ns);
329 net = ERR_PTR(-EINVAL);
336 struct net *copy_net_ns(unsigned long flags, struct net *old_net)
338 if (flags & CLONE_NEWNET)
339 return ERR_PTR(-EINVAL);
343 struct net *get_net_ns_by_fd(int fd)
345 return ERR_PTR(-EINVAL);
349 struct net *get_net_ns_by_pid(pid_t pid)
351 struct task_struct *tsk;
354 /* Lookup the network namespace */
355 net = ERR_PTR(-ESRCH);
357 tsk = find_task_by_vpid(pid);
359 struct nsproxy *nsproxy;
360 nsproxy = task_nsproxy(tsk);
362 net = get_net(nsproxy->net_ns);
367 EXPORT_SYMBOL_GPL(get_net_ns_by_pid);
369 static int __init net_ns_init(void)
371 struct net_generic *ng;
374 net_cachep = kmem_cache_create("net_namespace", sizeof(struct net),
378 /* Create workqueue for cleanup */
379 netns_wq = create_singlethread_workqueue("netns");
381 panic("Could not create netns workq");
384 ng = net_alloc_generic();
386 panic("Could not allocate generic netns");
388 rcu_assign_pointer(init_net.gen, ng);
390 mutex_lock(&net_mutex);
391 if (setup_net(&init_net))
392 panic("Could not setup the initial network namespace");
395 list_add_tail_rcu(&init_net.list, &net_namespace_list);
398 mutex_unlock(&net_mutex);
403 pure_initcall(net_ns_init);
406 static int __register_pernet_operations(struct list_head *list,
407 struct pernet_operations *ops)
411 LIST_HEAD(net_exit_list);
413 list_add_tail(&ops->list, list);
414 if (ops->init || (ops->id && ops->size)) {
416 error = ops_init(ops, net);
419 list_add_tail(&net->exit_list, &net_exit_list);
425 /* If I have an error cleanup all namespaces I initialized */
426 list_del(&ops->list);
427 ops_exit_list(ops, &net_exit_list);
428 ops_free_list(ops, &net_exit_list);
432 static void __unregister_pernet_operations(struct pernet_operations *ops)
435 LIST_HEAD(net_exit_list);
437 list_del(&ops->list);
439 list_add_tail(&net->exit_list, &net_exit_list);
440 ops_exit_list(ops, &net_exit_list);
441 ops_free_list(ops, &net_exit_list);
446 static int __register_pernet_operations(struct list_head *list,
447 struct pernet_operations *ops)
450 err = ops_init(ops, &init_net);
452 ops_free(ops, &init_net);
457 static void __unregister_pernet_operations(struct pernet_operations *ops)
459 LIST_HEAD(net_exit_list);
460 list_add(&init_net.exit_list, &net_exit_list);
461 ops_exit_list(ops, &net_exit_list);
462 ops_free_list(ops, &net_exit_list);
465 #endif /* CONFIG_NET_NS */
467 static DEFINE_IDA(net_generic_ids);
469 static int register_pernet_operations(struct list_head *list,
470 struct pernet_operations *ops)
476 error = ida_get_new_above(&net_generic_ids, 1, ops->id);
478 if (error == -EAGAIN) {
479 ida_pre_get(&net_generic_ids, GFP_KERNEL);
485 error = __register_pernet_operations(list, ops);
489 ida_remove(&net_generic_ids, *ops->id);
495 static void unregister_pernet_operations(struct pernet_operations *ops)
498 __unregister_pernet_operations(ops);
501 ida_remove(&net_generic_ids, *ops->id);
505 * register_pernet_subsys - register a network namespace subsystem
506 * @ops: pernet operations structure for the subsystem
508 * Register a subsystem which has init and exit functions
509 * that are called when network namespaces are created and
510 * destroyed respectively.
512 * When registered all network namespace init functions are
513 * called for every existing network namespace. Allowing kernel
514 * modules to have a race free view of the set of network namespaces.
516 * When a new network namespace is created all of the init
517 * methods are called in the order in which they were registered.
519 * When a network namespace is destroyed all of the exit methods
520 * are called in the reverse of the order with which they were
523 int register_pernet_subsys(struct pernet_operations *ops)
526 mutex_lock(&net_mutex);
527 error = register_pernet_operations(first_device, ops);
528 mutex_unlock(&net_mutex);
531 EXPORT_SYMBOL_GPL(register_pernet_subsys);
534 * unregister_pernet_subsys - unregister a network namespace subsystem
535 * @ops: pernet operations structure to manipulate
537 * Remove the pernet operations structure from the list to be
538 * used when network namespaces are created or destroyed. In
539 * addition run the exit method for all existing network
542 void unregister_pernet_subsys(struct pernet_operations *ops)
544 mutex_lock(&net_mutex);
545 unregister_pernet_operations(ops);
546 mutex_unlock(&net_mutex);
548 EXPORT_SYMBOL_GPL(unregister_pernet_subsys);
551 * register_pernet_device - register a network namespace device
552 * @ops: pernet operations structure for the subsystem
554 * Register a device which has init and exit functions
555 * that are called when network namespaces are created and
556 * destroyed respectively.
558 * When registered all network namespace init functions are
559 * called for every existing network namespace. Allowing kernel
560 * modules to have a race free view of the set of network namespaces.
562 * When a new network namespace is created all of the init
563 * methods are called in the order in which they were registered.
565 * When a network namespace is destroyed all of the exit methods
566 * are called in the reverse of the order with which they were
569 int register_pernet_device(struct pernet_operations *ops)
572 mutex_lock(&net_mutex);
573 error = register_pernet_operations(&pernet_list, ops);
574 if (!error && (first_device == &pernet_list))
575 first_device = &ops->list;
576 mutex_unlock(&net_mutex);
579 EXPORT_SYMBOL_GPL(register_pernet_device);
582 * unregister_pernet_device - unregister a network namespace netdevice
583 * @ops: pernet operations structure to manipulate
585 * Remove the pernet operations structure from the list to be
586 * used when network namespaces are created or destroyed. In
587 * addition run the exit method for all existing network
590 void unregister_pernet_device(struct pernet_operations *ops)
592 mutex_lock(&net_mutex);
593 if (&ops->list == first_device)
594 first_device = first_device->next;
595 unregister_pernet_operations(ops);
596 mutex_unlock(&net_mutex);
598 EXPORT_SYMBOL_GPL(unregister_pernet_device);
601 static void *netns_get(struct task_struct *task)
603 struct net *net = NULL;
604 struct nsproxy *nsproxy;
607 nsproxy = task_nsproxy(task);
609 net = get_net(nsproxy->net_ns);
615 static void netns_put(void *ns)
620 static int netns_install(struct nsproxy *nsproxy, void *ns)
622 put_net(nsproxy->net_ns);
623 nsproxy->net_ns = get_net(ns);
627 const struct proc_ns_operations netns_operations = {
629 .type = CLONE_NEWNET,
632 .install = netns_install,